Alibaba Unveils ‘Tongyi DeepResearch’, A 30B-Parameter AI Model that Outperforms GPT-4

Alibaba, a prominent e-commerce platform, has introduced ‘Tongyi DeepResearch’, a 30-billion-parameter agentic model that evidently outperforms GPT-4 and DeepSeek-V3 in deep abilities. Alibaba is activating only 3.3 billion parameters at runtime. Furthermore, the interesting news is that the model was trained on just two NVIDIA H100 GPUs. Spheron Network, known for building the world’s largest community-powered data center for AI workload, has revealed this news through its official X account.

This might be the biggest AI plot twist of 2025.

Alibaba has just unveiled Tongyi DeepResearch, a 30B-parameter agentic model that outperforms GPT-4 and DeepSeek-V3 in deep research, while activating only 3.3B parameters at runtime.

• Fully open-source.
• Trained on just 2… pic.twitter.com/BSaEYfaTmJ

— Spheron Network (@spheron) November 2, 2025

 According to Alibaba, Tongyi DeepResearch spots a major jump in compute-efficient artificial intelligence (AI). The model’s distinctive advance lies in a new “agentic mid-training phase,” where it acquires knowledge to reason, plan, and use tools before the traditional fine-tuning process.

Alibaba’s Open-Source Step toward Smarter AI Thinking

Tongyi DeepResearch is working differently to make its unique identity in the market. Therefore, Tongyi DeepResearch thinks like an agent to deliver more structured, desired outputs with minimum computational cost.

Since, completely built on synthetic data, the model has been trained on 32000+ token samples and executes 10+ tool calls per task, which is in itself a brilliant achievement and shows an advanced capability to tackle complex reasoning. On the other hand, Alibaba, in collaboration with the AI landscape to develop model fully open-source model, signaling a push toward greater accessibility.

Spheron Network Redefines the Future of AI Compute

Spheron Network declared that this is a decisive moment for the industry. This change brings the massive centralized data centers to decentralized and on-demand GPU access. All of this is happening with the aid of AI to make dreams true. In addition, Spheron emphasized that “compute efficiency, not compute dominance, and will define the next era of AI.”

With such achievement like Tongyi DeepResearch appearing, it’s becoming crystal clear that the competition in artificial intelligence is transitioning from only scale to intelligent efficiency. It is a decentralized innovation that brings the advanced wave of transformative AI technologies. After these advancements, it is possible to predict an open, efficient, and people-powered future.